Benchmark SGLASSO Screening Rules — benchmark_sglasso_screen

Runs a focused benchmark comparing "none", "SSR", and "SSR_fast" screening rules on block-correlated synthetic designs.

Usage

benchmark_sglasso_screen_rules(
  scenarios = data.frame(n = c(200L, 300L), p = c(20000L, 50000L), J = c(2000L, 5000L)),
  nlambda = 100,
  active_groups = 20,
  rho_within = 0.7,
  rho_between = 0.3,
  alpha = 0.5,
  d = 0.5,
  screens = c("none", "SSR", "SSR_fast"),
  reps = 1,
  seed = 2026,
  standardize = TRUE,
  eps = 1e-04,
  verbose = TRUE
)

Arguments

scenarios: A data frame with columns n, p, and J.
nlambda: Number of lambda values.
active_groups: Number of truly active groups.
rho_within: Within-group correlation.
rho_between: Between-group correlation.
alpha: Elastic net mixing parameter passed to sglasso.
d: Scale parameter passed to sglasso.
screens: Screening rules to compare.
reps: Number of repetitions per scenario.
seed: Random seed.
standardize: Whether sglasso should standardize internally.
eps: Convergence tolerance.
verbose: Print progress messages.

Value

A list containing raw run summaries, per-screen timing summaries, and pairwise accuracy comparisons against screen = "none".